lustre/mdt/mdt_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/mdt/mdt_lib.c
  37  *
  38  * Lustre Metadata Target (mdt) request unpacking helper.
  39  *
  40  * Author: Peter Braam <braam@clusterfs.com>
  41  * Author: Andreas Dilger <adilger@clusterfs.com>
  42  * Author: Phil Schwan <phil@clusterfs.com>
  43  * Author: Mike Shaver <shaver@clusterfs.com>
  44  * Author: Nikita Danilov <nikita@clusterfs.com>
  45  * Author: Huang Hua <huanghua@clusterfs.com>
  46  * Author: Fan Yong <fanyong@clusterfs.com>
  47  */
  48
  49 #define DEBUG_SUBSYSTEM S_MDS
  50
  51 #include <linux/user_namespace.h>
  52 #ifdef HAVE_UIDGID_HEADER
  53 # include <linux/uidgid.h>
  54 #endif
  55 #include "mdt_internal.h"
  56 #include <lnet/nidstr.h>
  57 #include <lustre_nodemap.h>
  58
  59 typedef enum ucred_init_type {
  60         NONE_INIT       = 0,
  61         BODY_INIT       = 1,
  62         REC_INIT        = 2
  63 } ucred_init_type_t;
  64
  65 void mdt_exit_ucred(struct mdt_thread_info *info)
  66 {
  67         struct lu_ucred   *uc  = mdt_ucred(info);
  68         struct mdt_device *mdt = info->mti_mdt;
  69
  70         LASSERT(uc != NULL);
  71         if (uc->uc_valid != UCRED_INIT) {
  72                 uc->uc_suppgids[0] = uc->uc_suppgids[1] = -1;
  73                 if (uc->uc_ginfo) {
  74                         put_group_info(uc->uc_ginfo);
  75                         uc->uc_ginfo = NULL;
  76                 }
  77                 if (uc->uc_identity) {
  78                         mdt_identity_put(mdt->mdt_identity_cache,
  79                                          uc->uc_identity);
  80                         uc->uc_identity = NULL;
  81                 }
  82                 uc->uc_valid = UCRED_INIT;
  83         }
  84 }
  85
  86 static int match_nosquash_list(struct rw_semaphore *sem,
  87                                struct list_head *nidlist,
  88                                lnet_nid_t peernid)
  89 {
  90         int rc;
  91         ENTRY;
  92         down_read(sem);
  93         rc = cfs_match_nid(peernid, nidlist);
  94         up_read(sem);
  95         RETURN(rc);
  96 }
  97
  98 /* root_squash for inter-MDS operations */
  99 static int mdt_root_squash(struct mdt_thread_info *info, lnet_nid_t peernid)
 100 {
 101         struct lu_ucred *ucred = mdt_ucred(info);
 102         struct root_squash_info *squash = &info->mti_mdt->mdt_squash;
 103         ENTRY;
 104
 105         LASSERT(ucred != NULL);
 106         if (!squash->rsi_uid || ucred->uc_fsuid)
 107                 RETURN(0);
 108
 109         if (match_nosquash_list(&squash->rsi_sem,
 110                                 &squash->rsi_nosquash_nids,
 111                                 peernid)) {
 112                 CDEBUG(D_OTHER, "%s is in nosquash_nids list\n",
 113                        libcfs_nid2str(peernid));
 114                 RETURN(0);
 115         }
 116
 117         CDEBUG(D_OTHER, "squash req from %s, (%d:%d/%x)=>(%d:%d/%x)\n",
 118                libcfs_nid2str(peernid),
 119                ucred->uc_fsuid, ucred->uc_fsgid, ucred->uc_cap,
 120                squash->rsi_uid, squash->rsi_gid, 0);
 121
 122         ucred->uc_fsuid = squash->rsi_uid;
 123         ucred->uc_fsgid = squash->rsi_gid;
 124         ucred->uc_cap = 0;
 125         ucred->uc_suppgids[0] = -1;
 126         ucred->uc_suppgids[1] = -1;
 127
 128         RETURN(0);
 129 }
 130
 131 static void ucred_set_jobid(struct mdt_thread_info *info, struct lu_ucred *uc)
 132 {
 133         struct ptlrpc_request   *req = mdt_info_req(info);
 134         const char              *jobid = mdt_req_get_jobid(req);
 135
 136         /* set jobid if specified. */
 137         if (jobid)
 138                 strlcpy(uc->uc_jobid, jobid, sizeof(uc->uc_jobid));
 139         else
 140                 uc->uc_jobid[0] = '\0';
 141 }
 142
 143 static int new_init_ucred(struct mdt_thread_info *info, ucred_init_type_t type,
 144                           void *buf)
 145 {
 146         struct ptlrpc_request   *req = mdt_info_req(info);
 147         struct mdt_device       *mdt = info->mti_mdt;
 148         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 149         struct lu_ucred         *ucred = mdt_ucred(info);
 150         lnet_nid_t               peernid = req->rq_peer.nid;
 151         __u32                    perm = 0;
 152         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 153         int                      setuid;
 154         int                      setgid;
 155         int                      rc = 0;
 156
 157         ENTRY;
 158
 159         LASSERT(req->rq_auth_gss);
 160         LASSERT(!req->rq_auth_usr_mdt);
 161         LASSERT(req->rq_user_desc);
 162         LASSERT(ucred != NULL);
 163
 164         ucred->uc_valid = UCRED_INVALID;
 165
 166         ucred->uc_o_uid   = pud->pud_uid;
 167         ucred->uc_o_gid   = pud->pud_gid;
 168         ucred->uc_o_fsuid = pud->pud_fsuid;
 169         ucred->uc_o_fsgid = pud->pud_fsgid;
 170
 171         if (type == BODY_INIT) {
 172                 struct mdt_body *body = (struct mdt_body *)buf;
 173
 174                 ucred->uc_suppgids[0] = body->mbo_suppgid;
 175                 ucred->uc_suppgids[1] = -1;
 176         }
 177
 178         /* sanity check: we expect the uid which client claimed is true */
 179         if (remote) {
 180                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 181                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 182                         RETURN(-EACCES);
 183                 }
 184
 185                 if (ptlrpc_user_desc_do_idmap(req, pud))
 186                         RETURN(-EACCES);
 187
 188                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 189                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 190                                "while client claims %u:%u/%u:%u\n",
 191                                libcfs_nid2str(peernid), req->rq_auth_uid,
 192                                req->rq_auth_mapped_uid,
 193                                pud->pud_uid, pud->pud_gid,
 194                                pud->pud_fsuid, pud->pud_fsgid);
 195                         RETURN(-EACCES);
 196                 }
 197         } else {
 198                 if (req->rq_auth_uid != pud->pud_uid) {
 199                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 200                                "while client claims %u:%u/%u:%u\n",
 201                                libcfs_nid2str(peernid), req->rq_auth_uid,
 202                                pud->pud_uid, pud->pud_gid,
 203                                pud->pud_fsuid, pud->pud_fsgid);
 204                         RETURN(-EACCES);
 205                 }
 206         }
 207
 208         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 209                 if (remote) {
 210                         CDEBUG(D_SEC, "remote client must run with identity_get "
 211                                "enabled!\n");
 212                         RETURN(-EACCES);
 213                 } else {
 214                         ucred->uc_identity = NULL;
 215                         perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 216                                CFS_SETGRP_PERM;
 217                 }
 218         } else {
 219                 struct md_identity *identity;
 220
 221                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 222                                             pud->pud_uid);
 223                 if (IS_ERR(identity)) {
 224                         if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 225                                      !remote)) {
 226                                 ucred->uc_identity = NULL;
 227                                 perm = CFS_SETUID_PERM | CFS_SETGID_PERM |
 228                                        CFS_SETGRP_PERM;
 229                         } else {
 230                                 CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 231                                        pud->pud_uid);
 232                                 RETURN(-EACCES);
 233                         }
 234                 } else {
 235                         ucred->uc_identity = identity;
 236                         perm = mdt_identity_get_perm(ucred->uc_identity,
 237                                                      remote, peernid);
 238                 }
 239         }
 240
 241         /* find out the setuid/setgid attempt */
 242         setuid = (pud->pud_uid != pud->pud_fsuid);
 243         setgid = ((pud->pud_gid != pud->pud_fsgid) ||
 244                   (ucred->uc_identity &&
 245                    (pud->pud_gid != ucred->uc_identity->mi_gid)));
 246
 247         /* check permission of setuid */
 248         if (setuid && !(perm & CFS_SETUID_PERM)) {
 249                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 250                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 251                 GOTO(out, rc = -EACCES);
 252         }
 253
 254         /* check permission of setgid */
 255         if (setgid && !(perm & CFS_SETGID_PERM)) {
 256                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 257                        "from %s\n", pud->pud_uid, pud->pud_gid,
 258                        pud->pud_fsuid, pud->pud_fsgid,
 259                        ucred->uc_identity->mi_gid, libcfs_nid2str(peernid));
 260                 GOTO(out, rc = -EACCES);
 261         }
 262
 263         /*
 264          * NB: remote client not allowed to setgroups anyway.
 265          */
 266         if (!remote && perm & CFS_SETGRP_PERM) {
 267                 if (pud->pud_ngroups) {
 268                         /* setgroups for local client */
 269                         ucred->uc_ginfo = groups_alloc(pud->pud_ngroups);
 270                         if (!ucred->uc_ginfo) {
 271                                 CERROR("failed to alloc %d groups\n",
 272                                        pud->pud_ngroups);
 273                                 GOTO(out, rc = -ENOMEM);
 274                         }
 275
 276                         lustre_groups_from_list(ucred->uc_ginfo,
 277                                                 pud->pud_groups);
 278                         lustre_groups_sort(ucred->uc_ginfo);
 279                 } else {
 280                         ucred->uc_ginfo = NULL;
 281                 }
 282         } else {
 283                 ucred->uc_suppgids[0] = -1;
 284                 ucred->uc_suppgids[1] = -1;
 285                 ucred->uc_ginfo = NULL;
 286         }
 287
 288         ucred->uc_uid   = pud->pud_uid;
 289         ucred->uc_gid   = pud->pud_gid;
 290         ucred->uc_fsuid = pud->pud_fsuid;
 291         ucred->uc_fsgid = pud->pud_fsgid;
 292
 293         /* process root_squash here. */
 294         mdt_root_squash(info, peernid);
 295
 296         /* remove fs privilege for non-root user. */
 297         if (ucred->uc_fsuid)
 298                 ucred->uc_cap = pud->pud_cap & ~CFS_CAP_FS_MASK;
 299         else
 300                 ucred->uc_cap = pud->pud_cap;
 301         if (remote && !(perm & CFS_RMTOWN_PERM))
 302                 ucred->uc_cap &= ~(CFS_CAP_SYS_RESOURCE_MASK |
 303                                    CFS_CAP_CHOWN_MASK);
 304         ucred->uc_valid = UCRED_NEW;
 305         ucred_set_jobid(info, ucred);
 306
 307         EXIT;
 308
 309 out:
 310         if (rc) {
 311                 if (ucred->uc_ginfo) {
 312                         put_group_info(ucred->uc_ginfo);
 313                         ucred->uc_ginfo = NULL;
 314                 }
 315                 if (ucred->uc_identity) {
 316                         mdt_identity_put(mdt->mdt_identity_cache,
 317                                          ucred->uc_identity);
 318                         ucred->uc_identity = NULL;
 319                 }
 320         }
 321
 322         return rc;
 323 }
 324
 325 /**
 326  * Check whether allow the client to set supplementary group IDs or not.
 327  *
 328  * \param[in] info      pointer to the thread context
 329  * \param[in] uc        pointer to the RPC user descriptor
 330  *
 331  * \retval              true if allow to set supplementary group IDs
 332  * \retval              false for other cases
 333  */
 334 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc)
 335 {
 336         __u32 remote = exp_connect_rmtclient(info->mti_exp);
 337         __u32 perm;
 338
 339         /* 1. If identity_upcall is disabled, then forbid remote client to set
 340          *    supplementary group IDs, but permit local client to do that. */
 341         if (is_identity_get_disabled(info->mti_mdt->mdt_identity_cache)) {
 342                 if (remote)
 343                         return false;
 344
 345                 return true;
 346         }
 347
 348         /* 2. If fail to get related identities, then forbid any client to
 349          *    set supplementary group IDs. */
 350         if (uc->uc_identity == NULL)
 351                 return false;
 352
 353         /* 3. Check the permission in the identities. */
 354         perm = mdt_identity_get_perm(uc->uc_identity, remote,
 355                                      mdt_info_req(info)->rq_peer.nid);
 356         if (perm & CFS_SETGRP_PERM)
 357                 return true;
 358
 359         return false;
 360 }
 361
 362 int mdt_check_ucred(struct mdt_thread_info *info)
 363 {
 364         struct ptlrpc_request   *req = mdt_info_req(info);
 365         struct mdt_device       *mdt = info->mti_mdt;
 366         struct ptlrpc_user_desc *pud = req->rq_user_desc;
 367         struct lu_ucred         *ucred = mdt_ucred(info);
 368         struct md_identity      *identity = NULL;
 369         lnet_nid_t               peernid = req->rq_peer.nid;
 370         __u32                    perm = 0;
 371         __u32                    remote = exp_connect_rmtclient(info->mti_exp);
 372         int                      setuid;
 373         int                      setgid;
 374         int                      rc = 0;
 375
 376         ENTRY;
 377
 378         LASSERT(ucred != NULL);
 379         if ((ucred->uc_valid == UCRED_OLD) || (ucred->uc_valid == UCRED_NEW))
 380                 RETURN(0);
 381
 382         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 383                 RETURN(0);
 384
 385         /* sanity check: if we use strong authentication, we expect the
 386          * uid which client claimed is true */
 387         if (remote) {
 388                 if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_mapped_uid))) {
 389                         CDEBUG(D_SEC, "remote user not mapped, deny access!\n");
 390                         RETURN(-EACCES);
 391                 }
 392
 393                 if (ptlrpc_user_desc_do_idmap(req, pud))
 394                         RETURN(-EACCES);
 395
 396                 if (req->rq_auth_mapped_uid != pud->pud_uid) {
 397                         CDEBUG(D_SEC, "remote client %s: auth/mapped uid %u/%u "
 398                                "while client claims %u:%u/%u:%u\n",
 399                                libcfs_nid2str(peernid), req->rq_auth_uid,
 400                                req->rq_auth_mapped_uid,
 401                                pud->pud_uid, pud->pud_gid,
 402                                pud->pud_fsuid, pud->pud_fsgid);
 403                         RETURN(-EACCES);
 404                 }
 405         } else {
 406                 if (req->rq_auth_uid != pud->pud_uid) {
 407                         CDEBUG(D_SEC, "local client %s: auth uid %u "
 408                                "while client claims %u:%u/%u:%u\n",
 409                                libcfs_nid2str(peernid), req->rq_auth_uid,
 410                                pud->pud_uid, pud->pud_gid,
 411                                pud->pud_fsuid, pud->pud_fsgid);
 412                         RETURN(-EACCES);
 413                 }
 414         }
 415
 416         if (is_identity_get_disabled(mdt->mdt_identity_cache)) {
 417                 if (remote) {
 418                         CDEBUG(D_SEC, "remote client must run with identity_get "
 419                                "enabled!\n");
 420                         RETURN(-EACCES);
 421                 }
 422                 RETURN(0);
 423         }
 424
 425         identity = mdt_identity_get(mdt->mdt_identity_cache, pud->pud_uid);
 426         if (IS_ERR(identity)) {
 427                 if (unlikely(PTR_ERR(identity) == -EREMCHG &&
 428                              !remote)) {
 429                         RETURN(0);
 430                 } else {
 431                         CDEBUG(D_SEC, "Deny access without identity: uid %u\n",
 432                                pud->pud_uid);
 433                         RETURN(-EACCES);
 434                }
 435         }
 436
 437         perm = mdt_identity_get_perm(identity, remote, peernid);
 438         /* find out the setuid/setgid attempt */
 439         setuid = (pud->pud_uid != pud->pud_fsuid);
 440         setgid = (pud->pud_gid != pud->pud_fsgid ||
 441                   pud->pud_gid != identity->mi_gid);
 442
 443         /* check permission of setuid */
 444         if (setuid && !(perm & CFS_SETUID_PERM)) {
 445                 CDEBUG(D_SEC, "mdt blocked setuid attempt (%u -> %u) from %s\n",
 446                        pud->pud_uid, pud->pud_fsuid, libcfs_nid2str(peernid));
 447                 GOTO(out, rc = -EACCES);
 448         }
 449
 450         /* check permission of setgid */
 451         if (setgid && !(perm & CFS_SETGID_PERM)) {
 452                 CDEBUG(D_SEC, "mdt blocked setgid attempt (%u:%u/%u:%u -> %u) "
 453                        "from %s\n", pud->pud_uid, pud->pud_gid,
 454                        pud->pud_fsuid, pud->pud_fsgid, identity->mi_gid,
 455                        libcfs_nid2str(peernid));
 456                 GOTO(out, rc = -EACCES);
 457         }
 458
 459         EXIT;
 460
 461 out:
 462         mdt_identity_put(mdt->mdt_identity_cache, identity);
 463         return rc;
 464 }
 465
 466 static void mdt_squash_nodemap_id(struct lu_ucred *ucred,
 467                                   struct lu_nodemap *nodemap)
 468 {
 469         if (ucred->uc_o_uid == nodemap->nm_squash_uid) {
 470                 ucred->uc_fsuid = nodemap->nm_squash_uid;
 471                 ucred->uc_fsgid = nodemap->nm_squash_gid;
 472                 ucred->uc_cap = 0;
 473                 ucred->uc_suppgids[0] = -1;
 474                 ucred->uc_suppgids[1] = -1;
 475         }
 476 }
 477
 478
 479 static int old_init_ucred(struct mdt_thread_info *info,
 480                           struct mdt_body *body)
 481 {
 482         struct lu_ucred         *uc = mdt_ucred(info);
 483         struct mdt_device       *mdt = info->mti_mdt;
 484         struct md_identity      *identity = NULL;
 485         struct lu_nodemap       *nodemap =
 486                 info->mti_exp->exp_target_data.ted_nodemap;
 487         ENTRY;
 488
 489         body->mbo_uid = nodemap_map_id(nodemap, NODEMAP_UID,
 490                                        NODEMAP_CLIENT_TO_FS, body->mbo_uid);
 491         body->mbo_gid = nodemap_map_id(nodemap, NODEMAP_GID,
 492                                        NODEMAP_CLIENT_TO_FS, body->mbo_gid);
 493         body->mbo_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 494                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsuid);
 495         body->mbo_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 496                                        NODEMAP_CLIENT_TO_FS, body->mbo_fsgid);
 497
 498         LASSERT(uc != NULL);
 499         uc->uc_valid = UCRED_INVALID;
 500         uc->uc_o_uid = uc->uc_uid = body->mbo_uid;
 501         uc->uc_o_gid = uc->uc_gid = body->mbo_gid;
 502         uc->uc_o_fsuid = uc->uc_fsuid = body->mbo_fsuid;
 503         uc->uc_o_fsgid = uc->uc_fsgid = body->mbo_fsgid;
 504         uc->uc_suppgids[0] = body->mbo_suppgid;
 505         uc->uc_suppgids[1] = -1;
 506         uc->uc_ginfo = NULL;
 507         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 508                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 509                                             uc->uc_fsuid);
 510                 if (IS_ERR(identity)) {
 511                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 512                                 identity = NULL;
 513                         } else {
 514                                 CDEBUG(D_SEC, "Deny access without identity: "
 515                                        "uid %u\n", uc->uc_fsuid);
 516                                 RETURN(-EACCES);
 517                         }
 518                 }
 519         }
 520         uc->uc_identity = identity;
 521
 522         mdt_squash_nodemap_id(uc, nodemap);
 523
 524         /* process root_squash here. */
 525         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 526
 527         /* remove fs privilege for non-root user. */
 528         if (uc->uc_fsuid)
 529                 uc->uc_cap = body->mbo_capability & ~CFS_CAP_FS_MASK;
 530         else
 531                 uc->uc_cap = body->mbo_capability;
 532         uc->uc_valid = UCRED_OLD;
 533         ucred_set_jobid(info, uc);
 534
 535         RETURN(0);
 536 }
 537
 538 static int old_init_ucred_reint(struct mdt_thread_info *info)
 539 {
 540         struct lu_ucred         *uc = mdt_ucred(info);
 541         struct mdt_device       *mdt = info->mti_mdt;
 542         struct md_identity      *identity = NULL;
 543         struct lu_nodemap       *nodemap =
 544                 info->mti_exp->exp_target_data.ted_nodemap;
 545         ENTRY;
 546
 547         LASSERT(uc != NULL);
 548
 549         uc->uc_fsuid = nodemap_map_id(nodemap, NODEMAP_UID,
 550                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsuid);
 551         uc->uc_fsgid = nodemap_map_id(nodemap, NODEMAP_GID,
 552                                       NODEMAP_CLIENT_TO_FS, uc->uc_fsgid);
 553
 554         uc->uc_valid = UCRED_INVALID;
 555         uc->uc_o_uid = uc->uc_o_fsuid = uc->uc_uid = uc->uc_fsuid;
 556         uc->uc_o_gid = uc->uc_o_fsgid = uc->uc_gid = uc->uc_fsgid;
 557         uc->uc_ginfo = NULL;
 558
 559         if (!is_identity_get_disabled(mdt->mdt_identity_cache)) {
 560                 identity = mdt_identity_get(mdt->mdt_identity_cache,
 561                                             uc->uc_fsuid);
 562                 if (IS_ERR(identity)) {
 563                         if (unlikely(PTR_ERR(identity) == -EREMCHG)) {
 564                                 identity = NULL;
 565                         } else {
 566                                 CDEBUG(D_SEC, "Deny access without identity: "
 567                                        "uid %u\n", uc->uc_fsuid);
 568                                 RETURN(-EACCES);
 569                         }
 570                 }
 571         }
 572         uc->uc_identity = identity;
 573
 574         /* process root_squash here. */
 575         mdt_root_squash(info, mdt_info_req(info)->rq_peer.nid);
 576
 577         /* remove fs privilege for non-root user. */
 578         if (uc->uc_fsuid)
 579                 uc->uc_cap &= ~CFS_CAP_FS_MASK;
 580         uc->uc_valid = UCRED_OLD;
 581         ucred_set_jobid(info, uc);
 582
 583         RETURN(0);
 584 }
 585
 586 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body)
 587 {
 588         struct ptlrpc_request *req = mdt_info_req(info);
 589         struct lu_ucred       *uc  = mdt_ucred(info);
 590
 591         LASSERT(uc != NULL);
 592         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 593                 return 0;
 594
 595         mdt_exit_ucred(info);
 596
 597         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 598                 return old_init_ucred(info, body);
 599         else
 600                 return new_init_ucred(info, BODY_INIT, body);
 601 }
 602
 603 int mdt_init_ucred_reint(struct mdt_thread_info *info)
 604 {
 605         struct ptlrpc_request *req = mdt_info_req(info);
 606         struct lu_ucred       *uc  = mdt_ucred(info);
 607
 608         LASSERT(uc != NULL);
 609         if ((uc->uc_valid == UCRED_OLD) || (uc->uc_valid == UCRED_NEW))
 610                 return 0;
 611
 612         mdt_exit_ucred(info);
 613
 614         if (!req->rq_auth_gss || req->rq_auth_usr_mdt || !req->rq_user_desc)
 615                 return old_init_ucred_reint(info);
 616         else
 617                 return new_init_ucred(info, REC_INIT, NULL);
 618 }
 619
 620 /* copied from lov/lov_ea.c, just for debugging, will be removed later */
 621 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid)
 622 {
 623         const struct lov_ost_data_v1    *lod;
 624         int                              i;
 625         __u16                            count;
 626
 627         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 628                 return;
 629
 630         count = le16_to_cpu(((struct lov_user_md *)lmm)->lmm_stripe_count);
 631
 632         CDEBUG(level, "objid "DOSTID", magic 0x%08X, pattern %#X\n",
 633                POSTID(&lmm->lmm_oi), le32_to_cpu(lmm->lmm_magic),
 634                le32_to_cpu(lmm->lmm_pattern));
 635         CDEBUG(level, "stripe_size=0x%x, stripe_count=0x%x\n",
 636                le32_to_cpu(lmm->lmm_stripe_size), count);
 637
 638         /* If it's a directory or a released file, then there are
 639          * no actual objects to print, so bail out. */
 640         if (valid & OBD_MD_FLDIREA ||
 641             le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_RELEASED)
 642                 return;
 643
 644         LASSERT(count <= LOV_MAX_STRIPE_COUNT);
 645         for (i = 0, lod = lmm->lmm_objects; i < count; i++, lod++) {
 646                 struct ost_id oi;
 647
 648                 ostid_le_to_cpu(&lod->l_ost_oi, &oi);
 649                 CDEBUG(level, "stripe %u idx %u subobj "DOSTID"\n",
 650                        i, le32_to_cpu(lod->l_ost_idx), POSTID(&oi));
 651         }
 652 }
 653
 654 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv)
 655 {
 656         const struct lmv_mds_md_v1 *lmm1;
 657         int                        i;
 658
 659         if (likely(!cfs_cdebug_show(level, DEBUG_SUBSYSTEM)))
 660                 return;
 661
 662         lmm1 = &lmv->lmv_md_v1;
 663         CDEBUG(level, "magic 0x%08X, master %#X stripe_count %#x\n",
 664                le32_to_cpu(lmm1->lmv_magic),
 665                le32_to_cpu(lmm1->lmv_master_mdt_index),
 666                le32_to_cpu(lmm1->lmv_stripe_count));
 667
 668         if (le32_to_cpu(lmm1->lmv_magic) == LMV_MAGIC_STRIPE)
 669                 return;
 670
 671         for (i = 0; i < le32_to_cpu(lmm1->lmv_stripe_count); i++) {
 672                 struct lu_fid fid;
 673
 674                 fid_le_to_cpu(&fid, &lmm1->lmv_stripe_fids[i]);
 675                 CDEBUG(level, "idx %u subobj "DFID"\n", i, PFID(&fid));
 676         }
 677 }
 678
 679 /* Shrink and/or grow reply buffers */
 680 int mdt_fix_reply(struct mdt_thread_info *info)
 681 {
 682         struct req_capsule *pill = info->mti_pill;
 683         struct mdt_body    *body;
 684         int                md_size, md_packed = 0;
 685         int                acl_size;
 686         int                rc = 0;
 687         ENTRY;
 688
 689         body = req_capsule_server_get(pill, &RMF_MDT_BODY);
 690         LASSERT(body != NULL);
 691
 692         if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE |
 693                                OBD_MD_LINKNAME))
 694                 md_size = body->mbo_eadatasize;
 695         else
 696                 md_size = 0;
 697
 698         acl_size = body->mbo_aclsize;
 699
 700         /* this replay - not send info to client */
 701         if (info->mti_spec.no_create) {
 702                 md_size = 0;
 703                 acl_size = 0;
 704         }
 705
 706         CDEBUG(D_INFO, "Shrink to md_size = %d cookie/acl_size = %d\n",
 707                md_size, acl_size);
 708 /*
 709             &RMF_MDT_BODY,
 710             &RMF_MDT_MD,
 711             &RMF_ACL, or &RMF_LOGCOOKIES
 712 (optional)  &RMF_CAPA1,
 713 (optional)  &RMF_CAPA2,
 714 (optional)  something else
 715 */
 716
 717         /* MDT_MD buffer may be bigger than packed value, let's shrink all
 718          * buffers before growing it */
 719         if (info->mti_big_lmm_used) {
 720                 /* big_lmm buffer may be used even without packing the result
 721                  * into reply, just for internal server needs */
 722                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
 723                         md_packed = req_capsule_get_size(pill, &RMF_MDT_MD,
 724                                                          RCL_SERVER);
 725
 726                 /* free big lmm if md_size is not needed */
 727                 if (md_size == 0 || md_packed == 0) {
 728                         info->mti_big_lmm_used = 0;
 729                 } else {
 730                         /* buffer must be allocated separately */
 731                         LASSERT(info->mti_attr.ma_lmm !=
 732                                 req_capsule_server_get(pill, &RMF_MDT_MD));
 733                         req_capsule_shrink(pill, &RMF_MDT_MD, 0, RCL_SERVER);
 734                 }
 735         } else if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER)) {
 736                 req_capsule_shrink(pill, &RMF_MDT_MD, md_size, RCL_SERVER);
 737         }
 738
 739         if (req_capsule_has_field(pill, &RMF_ACL, RCL_SERVER))
 740                 req_capsule_shrink(pill, &RMF_ACL, acl_size, RCL_SERVER);
 741         else if (req_capsule_has_field(pill, &RMF_LOGCOOKIES, RCL_SERVER))
 742                 req_capsule_shrink(pill, &RMF_LOGCOOKIES,
 743                                    acl_size, RCL_SERVER);
 744
 745         if (req_capsule_has_field(pill, &RMF_CAPA1, RCL_SERVER) &&
 746             !(body->mbo_valid & OBD_MD_FLMDSCAPA))
 747                 req_capsule_shrink(pill, &RMF_CAPA1, 0, RCL_SERVER);
 748
 749         if (req_capsule_has_field(pill, &RMF_CAPA2, RCL_SERVER) &&
 750             !(body->mbo_valid & OBD_MD_FLOSSCAPA))
 751                 req_capsule_shrink(pill, &RMF_CAPA2, 0, RCL_SERVER);
 752
 753         /*
 754          * Some more field should be shrinked if needed.
 755          * This should be done by those who added fields to reply message.
 756          */
 757
 758         /* Grow MD buffer if needed finally */
 759         if (info->mti_big_lmm_used) {
 760                 void *lmm;
 761
 762                 LASSERT(md_size > md_packed);
 763                 CDEBUG(D_INFO, "Enlarge reply buffer, need extra %d bytes\n",
 764                        md_size - md_packed);
 765                 rc = req_capsule_server_grow(pill, &RMF_MDT_MD, md_size);
 766                 if (rc) {
 767                         /* we can't answer with proper LOV EA, drop flags,
 768                          * the rc is also returned so this request is
 769                          * considered as failed */
 770                         body->mbo_valid &= ~(OBD_MD_FLDIREA | OBD_MD_FLEASIZE);
 771                         /* don't return transno along with error */
 772                         lustre_msg_set_transno(pill->rc_req->rq_repmsg, 0);
 773                 } else {
 774                         /* now we need to pack right LOV/LMV EA */
 775                         lmm = req_capsule_server_get(pill, &RMF_MDT_MD);
 776                         if (info->mti_attr.ma_valid & MA_LOV) {
 777                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 778                                                              RCL_SERVER) ==
 779                                                 info->mti_attr.ma_lmm_size);
 780                                 memcpy(lmm, info->mti_attr.ma_lmm,
 781                                        info->mti_attr.ma_lmm_size);
 782                         } else if (info->mti_attr.ma_valid & MA_LMV) {
 783                                 LASSERT(req_capsule_get_size(pill, &RMF_MDT_MD,
 784                                                              RCL_SERVER) ==
 785                                                 info->mti_attr.ma_lmv_size);
 786                                 memcpy(lmm, info->mti_attr.ma_lmv,
 787                                        info->mti_attr.ma_lmv_size);
 788                         }
 789                 }
 790                 /* update mdt_max_mdsize so clients will be aware about that */
 791                 if (info->mti_mdt->mdt_max_mdsize < info->mti_attr.ma_lmm_size)
 792                         info->mti_mdt->mdt_max_mdsize =
 793                                                     info->mti_attr.ma_lmm_size;
 794                 info->mti_big_lmm_used = 0;
 795         }
 796         RETURN(rc);
 797 }
 798
 799
 800 /* if object is dying, pack the lov/llog data,
 801  * parameter info->mti_attr should be valid at this point! */
 802 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *mo,
 803                            const struct md_attr *ma)
 804 {
 805         struct mdt_body       *repbody;
 806         const struct lu_attr *la = &ma->ma_attr;
 807         ENTRY;
 808
 809         repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY);
 810         LASSERT(repbody != NULL);
 811
 812         if (ma->ma_valid & MA_INODE)
 813                 mdt_pack_attr2body(info, repbody, la, mdt_object_fid(mo));
 814
 815         if (ma->ma_valid & MA_LOV) {
 816                 CERROR("No need in LOV EA upon unlink\n");
 817                 dump_stack();
 818         }
 819         repbody->mbo_eadatasize = 0;
 820
 821         RETURN(0);
 822 }
 823
 824 static __u64 mdt_attr_valid_xlate(__u64 in, struct mdt_reint_record *rr,
 825                                   struct md_attr *ma)
 826 {
 827         __u64 out;
 828
 829         out = 0;
 830         if (in & MDS_ATTR_MODE)
 831                 out |= LA_MODE;
 832         if (in & MDS_ATTR_UID)
 833                 out |= LA_UID;
 834         if (in & MDS_ATTR_GID)
 835                 out |= LA_GID;
 836         if (in & MDS_ATTR_SIZE)
 837                 out |= LA_SIZE;
 838         if (in & MDS_ATTR_BLOCKS)
 839                 out |= LA_BLOCKS;
 840         if (in & MDS_ATTR_ATIME_SET)
 841                 out |= LA_ATIME;
 842         if (in & MDS_ATTR_CTIME_SET)
 843                 out |= LA_CTIME;
 844         if (in & MDS_ATTR_MTIME_SET)
 845                 out |= LA_MTIME;
 846         if (in & MDS_ATTR_ATTR_FLAG)
 847                 out |= LA_FLAGS;
 848         if (in & MDS_ATTR_KILL_SUID)
 849                 out |= LA_KILL_SUID;
 850         if (in & MDS_ATTR_KILL_SGID)
 851                 out |= LA_KILL_SGID;
 852
 853         if (in & MDS_ATTR_FROM_OPEN)
 854                 rr->rr_flags |= MRF_OPEN_TRUNC;
 855         if (in & MDS_OPEN_OWNEROVERRIDE)
 856                 ma->ma_attr_flags |= MDS_OWNEROVERRIDE;
 857         if (in & MDS_ATTR_FORCE)
 858                 ma->ma_attr_flags |= MDS_PERM_BYPASS;
 859
 860         in &= ~(MDS_ATTR_MODE | MDS_ATTR_UID | MDS_ATTR_GID |
 861                 MDS_ATTR_ATIME | MDS_ATTR_MTIME | MDS_ATTR_CTIME |
 862                 MDS_ATTR_ATIME_SET | MDS_ATTR_CTIME_SET | MDS_ATTR_MTIME_SET |
 863                 MDS_ATTR_SIZE | MDS_ATTR_BLOCKS | MDS_ATTR_ATTR_FLAG |
 864                 MDS_ATTR_FORCE | MDS_ATTR_KILL_SUID | MDS_ATTR_KILL_SGID |
 865                 MDS_ATTR_FROM_OPEN | MDS_OPEN_OWNEROVERRIDE);
 866         if (in != 0)
 867                 CERROR("Unknown attr bits: "LPX64"\n", in);
 868         return out;
 869 }
 870
 871 /* unpacking */
 872
 873 int mdt_name_unpack(struct req_capsule *pill,
 874                     const struct req_msg_field *field,
 875                     struct lu_name *ln,
 876                     enum mdt_name_flags flags)
 877 {
 878         ln->ln_name = req_capsule_client_get(pill, field);
 879         ln->ln_namelen = req_capsule_get_size(pill, field, RCL_CLIENT) - 1;
 880
 881         if (!lu_name_is_valid(ln)) {
 882                 ln->ln_name = NULL;
 883                 ln->ln_namelen = 0;
 884
 885                 return -EPROTO;
 886         }
 887
 888         if ((flags & MNF_FIX_ANON) &&
 889             ln->ln_namelen == 1 && ln->ln_name[0] == '/') {
 890                 /* Newer (3.x) kernels use a name of "/" for the
 891                  * "anonymous" disconnected dentries from NFS
 892                  * filehandle conversion. See d_obtain_alias(). */
 893                 ln->ln_name = NULL;
 894                 ln->ln_namelen = 0;
 895         }
 896
 897         return 0;
 898 }
 899
 900 static int mdt_setattr_unpack_rec(struct mdt_thread_info *info)
 901 {
 902         struct lu_ucred         *uc = mdt_ucred(info);
 903         struct md_attr          *ma = &info->mti_attr;
 904         struct lu_attr          *la = &ma->ma_attr;
 905         struct req_capsule      *pill = info->mti_pill;
 906         struct mdt_reint_record *rr = &info->mti_rr;
 907         struct mdt_rec_setattr  *rec;
 908         struct lu_nodemap       *nodemap =
 909                 info->mti_exp->exp_target_data.ted_nodemap;
 910         ENTRY;
 911
 912         CLASSERT(sizeof(struct mdt_rec_setattr)== sizeof(struct mdt_rec_reint));
 913         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
 914         if (rec == NULL)
 915                 RETURN(-EFAULT);
 916
 917         /* This prior initialization is needed for old_init_ucred_reint() */
 918         uc->uc_fsuid = rec->sa_fsuid;
 919         uc->uc_fsgid = rec->sa_fsgid;
 920         uc->uc_cap   = rec->sa_cap;
 921         uc->uc_suppgids[0] = rec->sa_suppgid;
 922         uc->uc_suppgids[1] = -1;
 923
 924         rr->rr_fid1 = &rec->sa_fid;
 925         la->la_valid = mdt_attr_valid_xlate(rec->sa_valid, rr, ma);
 926         la->la_mode  = rec->sa_mode;
 927         la->la_flags = rec->sa_attr_flags;
 928         la->la_uid   = nodemap_map_id(nodemap, NODEMAP_UID,
 929                                       NODEMAP_CLIENT_TO_FS, rec->sa_uid);
 930         la->la_gid   = nodemap_map_id(nodemap, NODEMAP_GID,
 931                                       NODEMAP_CLIENT_TO_FS, rec->sa_gid);
 932         la->la_size  = rec->sa_size;
 933         la->la_blocks = rec->sa_blocks;
 934         la->la_ctime = rec->sa_ctime;
 935         la->la_atime = rec->sa_atime;
 936         la->la_mtime = rec->sa_mtime;
 937         ma->ma_valid = MA_INODE;
 938
 939         if (rec->sa_bias & MDS_DATA_MODIFIED)
 940                 ma->ma_attr_flags |= MDS_DATA_MODIFIED;
 941         else
 942                 ma->ma_attr_flags &= ~MDS_DATA_MODIFIED;
 943
 944         if (rec->sa_bias & MDS_HSM_RELEASE)
 945                 ma->ma_attr_flags |= MDS_HSM_RELEASE;
 946         else
 947                 ma->ma_attr_flags &= ~MDS_HSM_RELEASE;
 948
 949         RETURN(0);
 950 }
 951
 952 static int mdt_close_handle_unpack(struct mdt_thread_info *info)
 953 {
 954         struct req_capsule *pill = info->mti_pill;
 955         struct mdt_ioepoch *ioepoch;
 956         ENTRY;
 957
 958         if (req_capsule_get_size(pill, &RMF_MDT_EPOCH, RCL_CLIENT))
 959                 ioepoch = req_capsule_client_get(pill, &RMF_MDT_EPOCH);
 960         else
 961                 ioepoch = NULL;
 962
 963         if (ioepoch == NULL)
 964                 RETURN(-EPROTO);
 965
 966         info->mti_close_handle = ioepoch->mio_handle;
 967
 968         RETURN(0);
 969 }
 970
 971 static inline int mdt_dlmreq_unpack(struct mdt_thread_info *info) {
 972         struct req_capsule      *pill = info->mti_pill;
 973
 974         if (req_capsule_get_size(pill, &RMF_DLM_REQ, RCL_CLIENT)) {
 975                 info->mti_dlm_req = req_capsule_client_get(pill, &RMF_DLM_REQ);
 976                 if (info->mti_dlm_req == NULL)
 977                         RETURN(-EFAULT);
 978         }
 979
 980         RETURN(0);
 981 }
 982
 983 static int mdt_setattr_unpack(struct mdt_thread_info *info)
 984 {
 985         struct mdt_reint_record *rr = &info->mti_rr;
 986         struct md_attr          *ma = &info->mti_attr;
 987         struct req_capsule      *pill = info->mti_pill;
 988         int rc;
 989         ENTRY;
 990
 991         rc = mdt_setattr_unpack_rec(info);
 992         if (rc)
 993                 RETURN(rc);
 994
 995         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
 996                 rr->rr_eadata = req_capsule_client_get(pill, &RMF_EADATA);
 997                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
 998                                                         RCL_CLIENT);
 999                 if (rr->rr_eadatalen > 0) {
1000                         const struct lmv_user_md        *lum;
1001
1002                         lum = rr->rr_eadata;
1003                         /* Sigh ma_valid(from req) does not indicate whether
1004                          * it will set LOV/LMV EA, so we have to check magic */
1005                         if (le32_to_cpu(lum->lum_magic) == LMV_USER_MAGIC) {
1006                                 ma->ma_valid |= MA_LMV;
1007                                 ma->ma_lmv = (void *)rr->rr_eadata;
1008                                 ma->ma_lmv_size = rr->rr_eadatalen;
1009                         } else {
1010                                 ma->ma_valid |= MA_LOV;
1011                                 ma->ma_lmm = (void *)rr->rr_eadata;
1012                                 ma->ma_lmm_size = rr->rr_eadatalen;
1013                         }
1014                 }
1015         }
1016
1017         rc = mdt_dlmreq_unpack(info);
1018         RETURN(rc);
1019 }
1020
1021 static int mdt_hsm_release_unpack(struct mdt_thread_info *info)
1022 {
1023         struct md_attr          *ma = &info->mti_attr;
1024         struct req_capsule      *pill = info->mti_pill;
1025         ENTRY;
1026
1027         if (!(ma->ma_attr_flags & MDS_HSM_RELEASE))
1028                 RETURN(0);
1029
1030         req_capsule_extend(pill, &RQF_MDS_RELEASE_CLOSE);
1031
1032         if (!(req_capsule_has_field(pill, &RMF_CLOSE_DATA, RCL_CLIENT) &&
1033             req_capsule_field_present(pill, &RMF_CLOSE_DATA, RCL_CLIENT)))
1034                 RETURN(-EFAULT);
1035
1036         RETURN(0);
1037 }
1038
1039 int mdt_close_unpack(struct mdt_thread_info *info)
1040 {
1041         int rc;
1042         ENTRY;
1043
1044         rc = mdt_close_handle_unpack(info);
1045         if (rc)
1046                 RETURN(rc);
1047
1048         rc = mdt_setattr_unpack_rec(info);
1049         if (rc)
1050                 RETURN(rc);
1051
1052         rc = mdt_hsm_release_unpack(info);
1053         if (rc)
1054                 RETURN(rc);
1055
1056         RETURN(mdt_init_ucred_reint(info));
1057 }
1058
1059 static int mdt_create_unpack(struct mdt_thread_info *info)
1060 {
1061         struct lu_ucred         *uc  = mdt_ucred(info);
1062         struct mdt_rec_create   *rec;
1063         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1064         struct mdt_reint_record *rr = &info->mti_rr;
1065         struct req_capsule      *pill = info->mti_pill;
1066         struct md_op_spec       *sp = &info->mti_spec;
1067         int rc;
1068         ENTRY;
1069
1070         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1071         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1072         if (rec == NULL)
1073                 RETURN(-EFAULT);
1074
1075         /* This prior initialization is needed for old_init_ucred_reint() */
1076         uc->uc_fsuid = rec->cr_fsuid;
1077         uc->uc_fsgid = rec->cr_fsgid;
1078         uc->uc_cap   = rec->cr_cap;
1079         uc->uc_suppgids[0] = rec->cr_suppgid1;
1080         uc->uc_suppgids[1] = -1;
1081         uc->uc_umask = rec->cr_umask;
1082
1083         rr->rr_fid1 = &rec->cr_fid1;
1084         rr->rr_fid2 = &rec->cr_fid2;
1085         attr->la_mode = rec->cr_mode;
1086         attr->la_rdev  = rec->cr_rdev;
1087         attr->la_uid   = rec->cr_fsuid;
1088         attr->la_gid   = rec->cr_fsgid;
1089         attr->la_ctime = rec->cr_time;
1090         attr->la_mtime = rec->cr_time;
1091         attr->la_atime = rec->cr_time;
1092         attr->la_valid = LA_MODE | LA_RDEV | LA_UID | LA_GID | LA_TYPE |
1093                          LA_CTIME | LA_MTIME | LA_ATIME;
1094         memset(&sp->u, 0, sizeof(sp->u));
1095         sp->sp_cr_flags = get_mrc_cr_flags(rec);
1096
1097         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1098         if (rc < 0)
1099                 RETURN(rc);
1100
1101         if (S_ISLNK(attr->la_mode)) {
1102                 const char *tgt = NULL;
1103
1104                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_SYM);
1105                 if (req_capsule_get_size(pill, &RMF_SYMTGT, RCL_CLIENT)) {
1106                         tgt = req_capsule_client_get(pill, &RMF_SYMTGT);
1107                         sp->u.sp_symname = tgt;
1108                 }
1109                 if (tgt == NULL)
1110                         RETURN(-EFAULT);
1111         } else {
1112                 req_capsule_extend(pill, &RQF_MDS_REINT_CREATE_RMT_ACL);
1113                 if (S_ISDIR(attr->la_mode) &&
1114                     req_capsule_get_size(pill, &RMF_EADATA, RCL_CLIENT) > 0) {
1115                         sp->u.sp_ea.eadata =
1116                                 req_capsule_client_get(pill, &RMF_EADATA);
1117                         sp->u.sp_ea.eadatalen =
1118                                 req_capsule_get_size(pill, &RMF_EADATA,
1119                                                      RCL_CLIENT);
1120                         sp->sp_cr_flags |= MDS_OPEN_HAS_EA;
1121                 }
1122         }
1123
1124         rc = mdt_dlmreq_unpack(info);
1125         RETURN(rc);
1126 }
1127
1128 static int mdt_link_unpack(struct mdt_thread_info *info)
1129 {
1130         struct lu_ucred         *uc  = mdt_ucred(info);
1131         struct mdt_rec_link     *rec;
1132         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1133         struct mdt_reint_record *rr = &info->mti_rr;
1134         struct req_capsule      *pill = info->mti_pill;
1135         int rc;
1136         ENTRY;
1137
1138         CLASSERT(sizeof(struct mdt_rec_link) == sizeof(struct mdt_rec_reint));
1139         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1140         if (rec == NULL)
1141                 RETURN(-EFAULT);
1142
1143         /* This prior initialization is needed for old_init_ucred_reint() */
1144         uc->uc_fsuid = rec->lk_fsuid;
1145         uc->uc_fsgid = rec->lk_fsgid;
1146         uc->uc_cap   = rec->lk_cap;
1147         uc->uc_suppgids[0] = rec->lk_suppgid1;
1148         uc->uc_suppgids[1] = rec->lk_suppgid2;
1149
1150         attr->la_uid = rec->lk_fsuid;
1151         attr->la_gid = rec->lk_fsgid;
1152         rr->rr_fid1 = &rec->lk_fid1;
1153         rr->rr_fid2 = &rec->lk_fid2;
1154         attr->la_ctime = rec->lk_time;
1155         attr->la_mtime = rec->lk_time;
1156         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME;
1157
1158         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1159         if (rc < 0)
1160                 RETURN(rc);
1161
1162         rc = mdt_dlmreq_unpack(info);
1163
1164         RETURN(rc);
1165 }
1166
1167 static int mdt_unlink_unpack(struct mdt_thread_info *info)
1168 {
1169         struct lu_ucred         *uc  = mdt_ucred(info);
1170         struct mdt_rec_unlink   *rec;
1171         struct md_attr          *ma = &info->mti_attr;
1172         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1173         struct mdt_reint_record *rr = &info->mti_rr;
1174         struct req_capsule      *pill = info->mti_pill;
1175         int rc;
1176         ENTRY;
1177
1178         CLASSERT(sizeof(struct mdt_rec_unlink) == sizeof(struct mdt_rec_reint));
1179         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1180         if (rec == NULL)
1181                 RETURN(-EFAULT);
1182
1183         /* This prior initialization is needed for old_init_ucred_reint() */
1184         uc->uc_fsuid = rec->ul_fsuid;
1185         uc->uc_fsgid = rec->ul_fsgid;
1186         uc->uc_cap   = rec->ul_cap;
1187         uc->uc_suppgids[0] = rec->ul_suppgid1;
1188         uc->uc_suppgids[1] = -1;
1189
1190         attr->la_uid = rec->ul_fsuid;
1191         attr->la_gid = rec->ul_fsgid;
1192         rr->rr_fid1 = &rec->ul_fid1;
1193         rr->rr_fid2 = &rec->ul_fid2;
1194         attr->la_ctime = rec->ul_time;
1195         attr->la_mtime = rec->ul_time;
1196         attr->la_mode  = rec->ul_mode;
1197         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1198
1199         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1200         if (rc < 0)
1201                 RETURN(rc);
1202
1203         if (rec->ul_bias & MDS_VTX_BYPASS)
1204                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1205         else
1206                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1207
1208         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1209
1210         rc = mdt_dlmreq_unpack(info);
1211         RETURN(rc);
1212 }
1213
1214 static int mdt_rmentry_unpack(struct mdt_thread_info *info)
1215 {
1216         info->mti_spec.sp_rm_entry = 1;
1217         return mdt_unlink_unpack(info);
1218 }
1219
1220 static int mdt_rename_unpack(struct mdt_thread_info *info)
1221 {
1222         struct lu_ucred         *uc = mdt_ucred(info);
1223         struct mdt_rec_rename   *rec;
1224         struct md_attr          *ma = &info->mti_attr;
1225         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1226         struct mdt_reint_record *rr = &info->mti_rr;
1227         struct req_capsule      *pill = info->mti_pill;
1228         int rc;
1229         ENTRY;
1230
1231         CLASSERT(sizeof(struct mdt_rec_rename) == sizeof(struct mdt_rec_reint));
1232         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1233         if (rec == NULL)
1234                 RETURN(-EFAULT);
1235
1236         /* This prior initialization is needed for old_init_ucred_reint() */
1237         uc->uc_fsuid = rec->rn_fsuid;
1238         uc->uc_fsgid = rec->rn_fsgid;
1239         uc->uc_cap   = rec->rn_cap;
1240         uc->uc_suppgids[0] = rec->rn_suppgid1;
1241         uc->uc_suppgids[1] = rec->rn_suppgid2;
1242
1243         attr->la_uid = rec->rn_fsuid;
1244         attr->la_gid = rec->rn_fsgid;
1245         rr->rr_fid1 = &rec->rn_fid1;
1246         rr->rr_fid2 = &rec->rn_fid2;
1247         attr->la_ctime = rec->rn_time;
1248         attr->la_mtime = rec->rn_time;
1249         /* rename_tgt contains the mode already */
1250         attr->la_mode = rec->rn_mode;
1251         attr->la_valid = LA_UID | LA_GID | LA_CTIME | LA_MTIME | LA_MODE;
1252
1253         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1254         if (rc < 0)
1255                 RETURN(rc);
1256
1257         rc = mdt_name_unpack(pill, &RMF_SYMTGT, &rr->rr_tgt_name, 0);
1258         if (rc < 0)
1259                 RETURN(rc);
1260
1261         if (rec->rn_bias & MDS_VTX_BYPASS)
1262                 ma->ma_attr_flags |= MDS_VTX_BYPASS;
1263         else
1264                 ma->ma_attr_flags &= ~MDS_VTX_BYPASS;
1265
1266         info->mti_spec.no_create = !!req_is_replay(mdt_info_req(info));
1267
1268
1269         rc = mdt_dlmreq_unpack(info);
1270
1271         RETURN(rc);
1272 }
1273
1274 /*
1275  * please see comment above LOV_MAGIC_V1_DEF
1276  */
1277 static void mdt_fix_lov_magic(struct mdt_thread_info *info)
1278 {
1279         struct mdt_reint_record *rr = &info->mti_rr;
1280         struct lov_user_md_v1   *v1;
1281
1282         v1 = (void *)rr->rr_eadata;
1283         LASSERT(v1);
1284
1285         if (unlikely(req_is_replay(mdt_info_req(info)))) {
1286                 if (v1->lmm_magic == LOV_USER_MAGIC_V1) {
1287                         v1->lmm_magic = LOV_MAGIC_V1_DEF;
1288                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V1)) {
1289                         v1->lmm_magic = __swab32(LOV_MAGIC_V1_DEF);
1290                 } else if (v1->lmm_magic == LOV_USER_MAGIC_V3) {
1291                         v1->lmm_magic = LOV_MAGIC_V3_DEF;
1292                 } else if (v1->lmm_magic == __swab32(LOV_USER_MAGIC_V3)) {
1293                         v1->lmm_magic = __swab32(LOV_MAGIC_V3_DEF);
1294                 }
1295         }
1296 }
1297
1298 static int mdt_open_unpack(struct mdt_thread_info *info)
1299 {
1300         struct lu_ucred         *uc = mdt_ucred(info);
1301         struct mdt_rec_create   *rec;
1302         struct lu_attr          *attr = &info->mti_attr.ma_attr;
1303         struct req_capsule      *pill = info->mti_pill;
1304         struct mdt_reint_record *rr   = &info->mti_rr;
1305         struct ptlrpc_request   *req  = mdt_info_req(info);
1306         struct md_op_spec       *sp   = &info->mti_spec;
1307         ENTRY;
1308
1309         CLASSERT(sizeof(struct mdt_rec_create) == sizeof(struct mdt_rec_reint));
1310         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1311         if (rec == NULL)
1312                 RETURN(-EFAULT);
1313
1314         /* This prior initialization is needed for old_init_ucred_reint() */
1315         uc->uc_fsuid = rec->cr_fsuid;
1316         uc->uc_fsgid = rec->cr_fsgid;
1317         uc->uc_cap   = rec->cr_cap;
1318         uc->uc_suppgids[0] = rec->cr_suppgid1;
1319         uc->uc_suppgids[1] = rec->cr_suppgid2;
1320         uc->uc_umask = rec->cr_umask;
1321
1322         rr->rr_fid1   = &rec->cr_fid1;
1323         rr->rr_fid2   = &rec->cr_fid2;
1324         rr->rr_handle = &rec->cr_old_handle;
1325         attr->la_mode = rec->cr_mode;
1326         attr->la_rdev  = rec->cr_rdev;
1327         attr->la_uid   = rec->cr_fsuid;
1328         attr->la_gid   = rec->cr_fsgid;
1329         attr->la_ctime = rec->cr_time;
1330         attr->la_mtime = rec->cr_time;
1331         attr->la_atime = rec->cr_time;
1332         attr->la_valid = LA_MODE  | LA_RDEV  | LA_UID   | LA_GID |
1333                          LA_CTIME | LA_MTIME | LA_ATIME;
1334         memset(&info->mti_spec.u, 0, sizeof(info->mti_spec.u));
1335         info->mti_spec.sp_cr_flags = get_mrc_cr_flags(rec);
1336         /* Do not trigger ASSERTION if client miss to set such flags. */
1337         if (unlikely(info->mti_spec.sp_cr_flags == 0))
1338                 RETURN(-EPROTO);
1339
1340         info->mti_cross_ref = !!(rec->cr_bias & MDS_CROSS_REF);
1341
1342         mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, MNF_FIX_ANON);
1343
1344         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1345                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1346                                                         RCL_CLIENT);
1347                 if (rr->rr_eadatalen > 0) {
1348                         rr->rr_eadata = req_capsule_client_get(pill,
1349                                                                &RMF_EADATA);
1350                         sp->u.sp_ea.eadatalen = rr->rr_eadatalen;
1351                         sp->u.sp_ea.eadata = rr->rr_eadata;
1352                         sp->no_create = !!req_is_replay(req);
1353                         mdt_fix_lov_magic(info);
1354                 }
1355
1356                 /*
1357                  * Client default md_size may be 0 right after client start,
1358                  * until all osc are connected, set here just some reasonable
1359                  * value to prevent misbehavior.
1360                  */
1361                 if (rr->rr_eadatalen == 0 &&
1362                     !(info->mti_spec.sp_cr_flags & MDS_OPEN_DELAY_CREATE))
1363                         rr->rr_eadatalen = MIN_MD_SIZE;
1364         }
1365
1366         RETURN(0);
1367 }
1368
1369 static int mdt_setxattr_unpack(struct mdt_thread_info *info)
1370 {
1371         struct mdt_reint_record *rr     = &info->mti_rr;
1372         struct lu_ucred         *uc     = mdt_ucred(info);
1373         struct lu_attr          *attr   = &info->mti_attr.ma_attr;
1374         struct req_capsule      *pill   = info->mti_pill;
1375         struct mdt_rec_setxattr *rec;
1376         int                      rc;
1377         ENTRY;
1378
1379
1380         CLASSERT(sizeof(struct mdt_rec_setxattr) ==
1381                          sizeof(struct mdt_rec_reint));
1382
1383         rec = req_capsule_client_get(pill, &RMF_REC_REINT);
1384         if (rec == NULL)
1385                 RETURN(-EFAULT);
1386
1387         /* This prior initialization is needed for old_init_ucred_reint() */
1388         uc->uc_fsuid  = rec->sx_fsuid;
1389         uc->uc_fsgid  = rec->sx_fsgid;
1390         uc->uc_cap    = rec->sx_cap;
1391         uc->uc_suppgids[0] = rec->sx_suppgid1;
1392         uc->uc_suppgids[1] = -1;
1393
1394         rr->rr_opcode = rec->sx_opcode;
1395         rr->rr_fid1   = &rec->sx_fid;
1396         attr->la_valid = rec->sx_valid;
1397         attr->la_ctime = rec->sx_time;
1398         attr->la_size = rec->sx_size;
1399         attr->la_flags = rec->sx_flags;
1400
1401         rc = mdt_name_unpack(pill, &RMF_NAME, &rr->rr_name, 0);
1402         if (rc < 0)
1403                 RETURN(rc);
1404
1405         if (req_capsule_field_present(pill, &RMF_EADATA, RCL_CLIENT)) {
1406                 rr->rr_eadatalen = req_capsule_get_size(pill, &RMF_EADATA,
1407                                                         RCL_CLIENT);
1408                 if (rr->rr_eadatalen > 0) {
1409                         rr->rr_eadata = req_capsule_client_get(pill,
1410                                                                &RMF_EADATA);
1411                         if (rr->rr_eadata == NULL)
1412                                 RETURN(-EFAULT);
1413                 } else {
1414                         rr->rr_eadata = NULL;
1415                 }
1416         } else if (!(attr->la_valid & OBD_MD_FLXATTRRM)) {
1417                 CDEBUG(D_INFO, "no xattr data supplied\n");
1418                 RETURN(-EFAULT);
1419         }
1420
1421         if (mdt_dlmreq_unpack(info) < 0)
1422                 RETURN(-EPROTO);
1423
1424         RETURN(0);
1425 }
1426
1427
1428 typedef int (*reint_unpacker)(struct mdt_thread_info *info);
1429
1430 static reint_unpacker mdt_reint_unpackers[REINT_MAX] = {
1431         [REINT_SETATTR]  = mdt_setattr_unpack,
1432         [REINT_CREATE]   = mdt_create_unpack,
1433         [REINT_LINK]     = mdt_link_unpack,
1434         [REINT_UNLINK]   = mdt_unlink_unpack,
1435         [REINT_RENAME]   = mdt_rename_unpack,
1436         [REINT_OPEN]     = mdt_open_unpack,
1437         [REINT_SETXATTR] = mdt_setxattr_unpack,
1438         [REINT_RMENTRY]  = mdt_rmentry_unpack,
1439         [REINT_MIGRATE]  = mdt_rename_unpack,
1440 };
1441
1442 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op)
1443 {
1444         int rc;
1445         ENTRY;
1446
1447         memset(&info->mti_rr, 0, sizeof(info->mti_rr));
1448         if (op < REINT_MAX && mdt_reint_unpackers[op] != NULL) {
1449                 info->mti_rr.rr_opcode = op;
1450                 rc = mdt_reint_unpackers[op](info);
1451         } else {
1452                 CERROR("Unexpected opcode %d\n", op);
1453                 rc = -EFAULT;
1454         }
1455         RETURN(rc);
1456 }